Back to Blogging
It’s been a long time since I last blogged. These last 3 years have been a whirlwind for both my career and personal life. It’s funny to think I’m back where I started, trying to found startups. I’ve always known I’d want to be a founder again but I didn’t know how long it’d take to get back here. Now that I’m back, it’s a daunting task to find a problem and solution you want to commit to for the next 5-10 years. That being said I’ve been really interested in exploring two areas for ideas, Artificial Intelligence and Crypto.
I believe we are still in the infancy of crypto and what it means for the tech ecosystem. I like Bitcoin as a long standing store of value which isn’t correlated with stocks, gold, real estate, or bonds. I believe that smart contracts and decentralization will revolutionize the finance space. I don’t believe in 95% of the altcoins that are out there. Recently I’ve been really interested in margin lending and borrowing. The idea of earning between 8-15% interest by holding stablecoins seems completely absurd to me. This led me down the rabbit hole into looking up Maker DAI and USDC by Coinbase along with the lending market providers dy/dx, Compound, and Nuo. From there I’ve found all kinds of interesting decentralized finance (DeFi) startups using these primitives to build new startups on top of it. Maybe my next startup idea will be building on top of this new DeFi platforms but shifting sands in both Multi-Collateral DAI and Ethereum 2.0 makes it difficult to say what the future for Ethereum looks like. Things just move so quickly in the this space I think it’s both a blessing and a curse.
After working at AI2 and seeing all the amazing work going on in Computer Vision and Natural Language Processing, I can’t help but be bullish about the prospects of building an AI-driven company. The problem I think the space has always had though is finding the killer applications for it. The other is the difficulty in building the entire system that goes along with making a robust AI pipeline. Execution risk here is extremely high but if you can do it right, it has an inherently high built-in moat to protect your company. Also given the rising trend in accessible API’s on mobile platforms and mobile-first base models in Tensorflow and Pytorch, I think we’ll see an explosion of AI enabled apps when the ecosystem is easier to build on. That being said the core problem is still figuring out what problems AI can solve today. This is where I tend to struggle.
I’ve always had a passion for physical fitness and I was wondering how I could marry physical fitness and Computer Vision. The most obvious one for me was can we use something like 2D pose-estimation (which deep learning has helped in speed/accuracy) to make some type of self-guided fitness coach. Certainly it’s been tried in the past but even just getting the tensorflowjs pose estimation models working on iOS proved a considerable challenge. If you want to take that and then apply it to real-life coaching you’d have to train a model per exercise to make it accurate enough or figure out why the pose estimation doesn’t transfer well to iOS. I forked and fixed some of the python code here.
So I am excited to get back on the startup ride but knowing so much more since the last one has its advantages and disadvantages. The most interesting is I’m more discerning about what ideas to pursue however this causes me to be more indecisive and maybe if I just stuck with one I’d be more successful.